d-math Analysis II: Several Variables ETH Zürich

Federico Franceschini Problem set 13 FS 2024

If you don’t have a lot of time focus on the Problems marked with (♡), if they have many
similar subtasks just attempt one or two of them.

13.1. Standard problem. Solve the initial value problem

y ′′ (x) + 4y ′ (x) + 4y(x) = 2e−2x , y(0) = y ′ (0) = 0.

“Solve” means two things: 1) write down a solution/a family of solutions and check that
indeed they work, 2) explain why such solution/family of solutions is unique (you may
refer to the notes).

13.2. Constant coefficient ODEs ♡. Solve the following differential equations, find
first the most general solution and then impose the initial conditions.
1. 2y ′′ − 24y ′ + 72y = 0, y(0) = 1, y ′ (0) = 4,
2. y ′′ + y ′ − 6y = 0, y(0) = 1, y ′ (0) = 7,
3. y ′′ − 2y ′ + 2y = 0, y(0) = 0, y ′ (0) = −1.
What changes if we allow complex-valued solutions?

13.3. Non-homogeneous problems ♡.

1. Determine the general solution of the differential equation

y ′′ − 4y ′ + 4y = sin(x).

2. Determine the solution of the initial value problem

y ′′ − y = x, y(0) = 1, y ′ (0) = 3.

3. Find the maximally defined solution of the initial value problem

4
 
′
y − + 1 y = x4 , y(1) = 1.
x
Hint: use the integrating factor trick. That is, look at the equation solved by
u(t) := eH(t) y(t), with H satisfying H ′ (t) = 4t + 1.

13.4. Cauchy-Lipschitz ♡. Put these differential equations in normal form, i.e. find in
each case U ⊂ R2 and F : U → R such that

y ′ (x) = F (x, y(x)), (x, y) ∈ U ⊂ R2 .

Then explain in each case whether Cauchy-Lipschitz-Picard theorem applies.

Finally find an explicit formula for the solution y(x) = ... for a general initial datum.
1. (x2 − x)y ′ = y 2 + y,

assignment: May 20, 2023 due: May 27, 2023 1/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

2. y ′ + ey = 1,
3. xy ′ = 1 − y 2 ,
4. y ′ x2 = y 2 + yx + x2 .

13.5. Multiple choice. Check the correct answer(s).

1. For which of the following differential equations does the set of all solutions form a
vector space?
(a) y ′ − 2y = 0.
(b) x(y ′ )2 − 2y = 0.
q
(c) y ′ − x |y| = 0.
(d) y ′ − y = 1.
(e) y ′ − x2 y = 0.
2. Let A ∈ Matn,n (R) be a real n × n matrix. Which of the following statements about
the solutions of the linear differential equation system ẋ = Ax are generally true?
(a) Every solution x : R → Rn with limt→∞ ∥x(t)∥2 = ∞ grows exponentially as
t → ∞.
(b) There exists a nontrivial constant solution x : R → Rn if and only if A has a
nontrivial kernel.
(c) If A is real diagonalizable and there exists a nontrivial bounded solution
x : R → Rn , then 0 is an eigenvalue of A.
(d) If A is complex diagonalizable and there exists a nontrivial bounded solution
x : R → Rn , then 0 is an eigenvalue of A.
3. Let U ⊂ R × Rn be open, f : U → Rn locally Lipschitz continuous in space and
(t0 , x0 ) ∈ U . Then the initial value problem

ẋ(t) = f (t, x(t)), x(t0 ) = x0 ,

has, by the Picard–Lindelöf theorem, there is an open interval Imax ⊂ R and a

maximal solution x : Imax → Rn whose graph is contained in U . Which of the
following statements generally hold for the Picard iteration for this initial value
problem?

(a) There exists a δ > 0 such that the Picard iteration on I = [t0 − δ, t0 + δ]
converges uniformly to x|I .
(b) The Picard iteration converges uniformly to x|Imax on Imax .
(c) The Picard iteration converges uniformly to x|K on every compact subinterval
K of Imax .

assignment: May 20, 2023 due: May 27, 2023 2/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

13.6. True or False ♡. Say whether the following statements are true or false:
1. Let A(t) ∈ C 1 (R, Rn×n ) be a time-dependent symmetric matrix and x0 ∈ Rn .
Then x(t) := exp A(t)x0 solves the ODE

x′ (t) := A′ (t)x(t), x(0) = 0?

2. Let A(t), B(t) ∈ C 1 (R, Rn×n ) be a time-dependent matrices. Then

(AB)′ (t) = A′ (t)B(t) + A(t)B ′ (t).

3. Let A(t), B(t) ∈ C 1 (R, Rn×n ) be time-dependent matrices. Then

exp(A(t) + B(t)) = exp(A(t)) · exp(B(t)).

4. Let A, B ∈ Rn×n be matrices, which are symmetric and positive definite. Then

exp(A + B) = exp(A) · exp(B).

assignment: May 20, 2023 due: May 27, 2023 3/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

13. Solutions

Solution of 13.1: The corresponding homogeneous equation has the characteristic

polynomial (λ + 2)2 and thus the general solution y0 (x) = (a + bx)e−2x . For the initial
value problem, the Ansatz y(x) = cx2 e−2x with the derivatives

y ′ (x) = (2cx − 2cx2 )e−2x ,

y ′′ (x) = (2c − 8cx + 4cx2 )e−2x .

Substituting into the inhomogeneous equation gives c = 1.

Solution of 13.2: The solutions are as in equations 7.15 to 7.17 in the script
1. The characteristic polynomial (divided by 2) is

p(X) = X 2 − 12X + 36 = (X − 6)(X − 6)

and due to the double root 6, the solutions are of the form

y(x) = C1 e6x + C2 xe6x

where C1 , C2 are in R or in C, depending on the desired solution set. With y(0) = 1

it follows C1 = 1 and with y ′ (0) = 4 it follows C2 = −2, so

y(x) = e6x − 2xe6x .

2. The characteristic polynomial is

p(X) = X 2 + X − 6 = (X + 3)(X − 2)

and thus the solutions are of the form

y(x) = C1 e−3x + C2 e2x

where C1 , C2 are in R or in C, depending on the desired solution set. With y(0) = 1

and y ′ (0) = −2 it follows C1 +C2 = 1 and −3C1 +2C2 = 7 it follows C1 = −1, C2 = 2,
so
y(x) = −e−3x + 2e2x .

3. The characteristic polynomial is

p(X) = X 2 + 2X + 2 = (X − (1 − i))(X − (1 + i))

and thus the complex solutions are of the form

y(x) = C1 e(1−i)x + C2 e(1+i)x

assignment: May 20, 2023 due: May 27, 2023 4/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

where C1 , C2 are in C. From

y(x) = ex (C1 e−ix + C2 eix )

= ex (C1 (cos x + i sin x) + C2 (cos x − i sin x))
= ex ((C1 + C2 ) cos x + (C1 − C2 )i sin x)

we can define Ce1 = C1 + C2 and Ce2 = (C1 − C2 )i and obtain

y(x) = ex (Ce1 cos x + Ce2 sin x)

(compare with chapter 14.2.1 in the script). Now we can demand C1 , C2 ∈ R and
obtain the real solutions.
With y(0) = 0 it follows Ce1 = 0 and with y ′ (0) = −1 it follows Ce2 = −1, so

y(x) = −ex sin x.

Solution of 13.3:
• First, we determine the homogeneous solution, i.e., the solution of

yh′′ − 4yh′ + 4yh = 0.

The characteristic polynomial is p(X) = X 2 − 4X + 4 = (X − 2)2 . Hence,

yh (x) = (A + Bx)e2x

for A, B ∈ C. For the particular solution, we assume yp (x) = C sin(x) + D cos(x).

Thus,

yp′ (x) = C cos(x) − D sin(x) and yp′ (x) = −C sin(x) − D cos(x)

Substituting yp , yp′ , and yp′′ into y ′′ − 4y ′ + 4y = sin(x), we obtain

(3C + 4D) sin(x) + (−4C + 3D) cos(x) = sin(x)

Since sin and cos are linearly independent, by comparing coefficients, we have

3 4 1
! ! !
C
=
−4 3 D 0

then !−1
3 4 1 1 3 −4 1 1 3
! ! ! ! !
C
= = =
D −4 3 0 25 4 3 0 25 4
thus our particular solution is yp (x) = 1
25
(3 sin(x) + 4 cos(x)). We conclude that
1
y(x) = yh (x) + yp (x) = (A + Bx)e2x + (3 sin(x) + 4 cos(x))
25
with A, B ∈ C is the general solution of our differential equation.

assignment: May 20, 2023 due: May 27, 2023 5/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

• First, we determine the homogeneous solution, i.e., the solution of

yh′′ − yh = 0.

The characteristic polynomial is p(X) = X 2 − 1. Hence,

yh (x) = Aex + Be−x

for A, B ∈ C. Since the right side of our differential equation is x, we assume

yp = Cx, C ∈ C, for the particular solution (see p.403 in the script). Thus,
yp′ (x) = C and yp′′ (x) = 0. Substituting yp , yp′ , and yp′′ into y ′′ − y = 0, we obtain
−Cx = x, i.e., C = −1. Thus, the general solution is

y(x) = Aex + Be−x − x

We now solve the initial value problem. From y(0) = 1, we obtain A + B = 1, and
from y ′ (0) = 3, we obtain A − B = 4. Thus, A = 52 and B = −3
2
. The solution to
the initial value problem is then
5 3
y(x) = ex − e−x − x.
2 2

• First, we determine the homogeneous solution, i.e., the solution of

4
 
′
y − +1 y =0
x
Since we want to solve the initial value problem y(1) = 1 in the end, we assume
x > 0. This differential equation is separable. We have
y ′ (x) 4
− =1+
y(x) x

from which we obtain y(x) = Aex x4 . Since the right side of our differential equation
is x4 , we assume yp = Cx4 + Dx3 + Ex2 + F x + G, C, D, E, F, G ∈ C, for the
particular solution (see p.403 in the script). Thus,

yp′ (x) = 4Cx3 + 3Dx2 + 2Ex + F

We obtain
4
 
(4Cx + 3Dx + 2Ex + F ) −
3 2
+ 1 (Cx4 + Dx3 + Ex2 + F x + G) = x4
x
By comparing coefficients, G = F = E = D = 0 and from
4
 
4Cx3 − + 1 Cx4 = x4
x
we obtain −Cx4 = x4 and finally C = −1. Thus, the general solution is

y(x) = x4 (Ae−x − 1)

assignment: May 20, 2023 due: May 27, 2023 6/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

We now solve the initial value problem. From y(1) = 1, we obtain

Ae − 1 = 1

i.e. A = 2e . The solution to the initial value problem is then

x4 (2ex − e)
y(x) = .
e

Solution of 13.4:
1.
y2 + y
U := {(x, y) : x ̸= 0, x ̸= 1}, F (x, y) := .
x2 − x
Since F ∈ C 1 (U ), Cauchy-Lipschitz applies. We separate the variables and obtain
y′ 1
= ,
y2 + y x2 − x
which is the same as
y′ y′ 1 1
− = − ,
y y+1 x−1 x
and after integrating with respect to x, we obtain
y x−1
log = log + C.
y+1 x
with an arbitrary constant C. It follows that
y x−1
= C̃ ,
y+1 x

where the two possibilities Ce = eC or Ce = −eC arise from removing the absolute
value. We solve for y:

yx = C̃(x − 1)(y + 1)
C̃(x − 1)
y(x) = .
x − C̃(x − 1)

2.
U := R2 , F (x, y) := 1 − ey .
Since F ∈ C 1 (U ), Cauchy-Lipschitz applies. We separate the variables and obtain
y′
= 1.
1 − ey
The right side integrates (with substitution ey = z, dz/dy = ey = z) as
ˆ ˆ ˆ ˆ
1 1 1 1 z z−1
dy = dz = dz + dz = log = − log
1−e y (1 − z)z 1−z z z−1 z

assignment: May 20, 2023 due: May 27, 2023 7/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

and therefore
ey − 1
log = −x − C
ey
with an arbitrary constant C. We solve for y:
ey − 1
= C̃e−x
ey
1
ey =
1 − C̃e−x
y(x) = − log(1 − C̃e−x ),

where C̃ is defined as in (a).

3.
1 − y2
U := {(x, y) : x ̸= 0}, F (x, y) := .
x
Since F ∈ C 1 (U ), Cauchy-Lipschitz applies. We separate the variables and obtain

y′ 1
= ,
1−y 2 x
which is the same as
1 y′ y′ 1
!
− = ,
2 y+1 y−1 x
and after integrating with respect to x, we obtain

1 y+1
log = log |x| + C
2 y−1

with an arbitrary constant C. We rearrange:

1 y+1
log = log |x| + C
2 y−1
y+1
log = log(|x|2 ) + 2C = log(x2 ) + 2C
y−1

and solve for y:

y+1
= C̃x2
y−1
C̃x2 + 1
y(x) = ,
C̃x2 − 1

where C̃ = ±e2C is defined analogously to (a).

4.
y 2 + xy + x2
U := {(x, y) : x ̸= 0}, F (x, y) := .
x2

assignment: May 20, 2023 due: May 27, 2023 8/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

Since F ∈ C 1 (U ), Cauchy-Lipschitz applies. We cannot directly separate the

variables. We write the differential equation as
 2
y y
y =
′
+ + 1.
x x

We use the substitution z(x) = y(x)

x
for x ̸= 0. Note that

y ′ (x)x − y(x)
z ′ (x) = ,
x2
i.e.
y ′ = xz ′ + z.
We obtain the differential equation

xz ′ + z = z 2 + z + 1.
z′
We separate the variables z 2 +1
= 1
x
and integrate with respect to x

arctan(z) = ln |x| + C

for an arbitrary constant C, i.e.

z = tan(ln |x| + C).

Returning to y:
y(x) = x tan(ln |x| + C).

Solution of 13.5:
(1a) Yes.
(1b) No.
(1c) No.
(1d) Yes.
(1e) Yes.
0 1 0 t
! !
(2a) No, take A = , then eAt = so if x(0) = e2 then |x(t)| = t goes to
0 0 0 0
infinity less than exponentially.
(2b) Yes, as 0 = x′ (t) = Ax(t) = Ax(0), so x(0) lies in the null space of A.
   
(2d) Counterexample: n = 2, A = 01 −1 0 , x(t) = sin(t) . Then x is a nontrivial bounded
cos(t)

solution and the eigenvalues of A are ±i. Since these are distinct, A is in particular
complex diagonalizable. Note that the argument in (c) implies that an eigenvalue
with real part 0 must exist.

assignment: May 20, 2023 due: May 27, 2023 9/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

(2c) If λ1 , . . . , λn ∈ R are the eigenvalues with corresponding basis v1 , . . . , vn ∈ Rn of

eigenvectors, then the general solution is given by
x(t) = c1 eλ1 t v1 + · · · + cn eλn t
for arbitrary c1 , . . . , cn ∈ R. (Indeed, the functions t 7→ eλj t vj for 1 ≤ j ≤ n are
solutions, and according to problem 4, they are also linearly independent. Since the
solution space is n-dimensional, cf. Proposition 14.15 and/or 14.40, these functions
thus form a basis of the solution space.) Due to the equivalence of all norms on Rn
(cf. Exercise 9.71), it follows for a constant C > 0 that
n o
∥x(t)∥2 ≥ C max |c1 |eλ1 t , . . . , |cn |eλn t .
Thus, if 0 is not among the eigenvalues and one of the values cj is not 0, ∥x(t)∥2
tends to ∞ either for t → ∞ or for t → −∞.
(3a) Yes.
(3b) No, consider x′ (t) = x(t), so that Imax = R. It is easy to check that each Picard
iteration is a polynomial function, which cannot be uniformly close to et on the
whole R.
• Yes. The graph of xK is compactly contained in U , so in that subregion f is globally
Lipschitz and bounded. Thus the Picard iterations restricted t0 K, that is xN |K are
equicontinuous. By Ascoli-Arzelá they converge uniformly to some function which
is necessarily the solution x|K .

Solution of 13.6:
1. False. It is not true, in general, that (exp A(t))′ = A′ (t) exp A(t). That something
must be wrong in that formula can be guessed indirectly noticing that in general
A′ (t) exp A(t) ̸= exp A(t)A′ (t).

2. True, as the product function · : Rn×n × Rn×n → Rn×n is a bilinear map. Here’s the
full-lenght proof: We write
A(t) = (aij (t))i,j and B(t) = (bij (t))i,j
where

aij, bij : [a, b] → R are differentiable functions. Then for example d
dt
A(t) =
d
a (t) . One recalls from Linear Algebra that
dt ij
m
(A(t)B(t))ij = aik (t)bkj (t)
X

k=1

It then holds
  m   m  
(A(t)B(t)) = a (t) bkj (t) + aik (t) b (t) =
X X
d d d
dt ij dt ik dt kj
k=1 k=1
     
= d
dt
A(t) B(t) + A(t) d
dt
B(t)
ij ij
     
and thus d
dt
A(t)B(t) = d
dt
A(t) B(t) + A(t) d
dt
B(t) .

assignment: May 20, 2023 due: May 27, 2023 10/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

3. False, see the next.

4. False, here’s a counterexample.
Consider the symmetric matrices:

1 0 0 1
! !
A= , B=
0 0 1 0

Let’s calculate exp(A).

Since A is a simple diagonal matrix, its exponential is straightforward:

1 0 e 0
!! !
exp(A) = exp =
0 0 0 1

Let’s calculate exp(B).

For B,
0 1 0 1 1 0
! ! !
B =2
= =I
1 0 1 0 0 1

Thus,
cosh(1) sinh(1)
!
exp(B) = I cosh(1) + B sinh(1) =
sinh(1) cosh(1)

Let’s calculate exp(A + B).

First, compute A + B:
1 1
!
A+B =
1 0

To compute the exponential of A+B, we need to find its eigenvalues and eigenvectors.
The characteristic polynomial of A + B is:

1−λ 1
!
det = λ2 − λ − 1
1 −λ

The eigenvalues are: √

1± 5
λ=
2
√ √
Let λ1 = 1+ 5
2
and λ2 = 1− 5
2
.
Using the eigenvalues, we can write:

eλ1 0
!
exp(A + B) = P P −1
0 eλ2

assignment: May 20, 2023 due: May 27, 2023 11/12

d-math Analysis II: Several Variables ETH Zürich
Federico Franceschini Problem set 13 FS 2024

where P is the matrix of eigenvectors of A + B. Finding P and P −1 explicitly is

more involved, but unnecessary. Indeed we compute exp(A) exp(B):

e 0 cosh(1) sinh(1) e cosh(1) e sinh(1)

! ! !
exp(A) exp(B) = =
0 1 sinh(1) cosh(1) sinh(1) cosh(1)

In order to show exp(A + B) ̸= exp(A) exp(B) it is enough to show that they have
different traces (for which the conjugation by P is harmless)

5.7 ∼ eλ1 + eλ2 = Tr(exp(A + B)) ̸= Tr(exp(A) exp(B)) = (e + 1) cosh(1) ∼ 5.5.

assignment: May 20, 2023 due: May 27, 2023 12/12